Fix #246086 (out of memory when building huge projects - don't read document contents when adding indexing job into queue, do it on demand)